Physics-Informed Neural Network-Based Nonlinear Model Predictive Control for Automated Guided Vehicle Trajectory Tracking

This paper proposes a nonlinear Model Predictive Control (MPC) method based on Physics-Informed Neural Networks (PINNs), aimed at enhancing the trajectory tracking performance of Automated Guided Vehicles (AGVs) in complex dynamic environments. Traditional physical models often face the challenges of computational inefficiency and insufficient control precision when dealing with complex dynamic systems. However, by integrating physical laws directly into the training process of neural networks, PINNs can effectively learn and capture the kinematic characteristics of vehicles, replacing traditional nonlinear ordinary differential equation models and thus significantly enhancing computational efficiency and control performance. During the model-training phase, this study further incorporates the Theory of Functional Connections (TFC) and adaptive loss balancing strategies to efficiently solve ODE problems without relying on numerical integration and optimize the control strategy. This combined approach not only reduces computational complexity, but also improves the robustness and precision of the control strategy in varying environments. Numerical simulations demonstrate that this method offers significant advantages in AGV trajectory-tracking tasks, manifested in higher computational efficiency and precise control performance. The proposal of the PINN-MPC method provides new theoretical support and innovative methods for real-time complex system control, with important research and application potential, and is expected to play a key role in future intelligent control systems.